Predicting risk in sentiment analysis using machine learning

Abstract

The purpose of this research is to do risk modeling after a sentiment analysis of Twitter posts based on a particular or certain sentiment with the help of the PRISM model.The model is named PRISM as the results obtained are an amalgamation of seven different attributes used in the research for comparison and tabulation of quantitative scores. These attributes are Accuracy, Precision, Recall, F1-Score, Support, Confusion Matrix, and Tweets. PRISM model can serve the law enforcement agencies in many ways and help them maintain peace, law and order in society as it is a proactive model. The sub-modules which are part of the PRISM model help to give quantitative values to predict the risk level on the sentiment of interest. After analysis of obtained testing results, it is observed that Support Vector Machine gives better results in accuracy, precision, F1-Score, Support and Recall as compared to the other three classifier models i.e. Naive Bayes, Decision Tree, and K nearest neighbor. It is also observed that with an increase or decrease in data, regarding the number of tweets, the fluctuation in performance of SVM is most stable i.e. it shows the least deviation and variation. The other algorithms show a considerable deviation in their performance.